This is the most complete explanation on the supposed "memory effect" that I have seen. It has been verified with a number for documents (ex. GE's tests) and other people. Please read it first, then we'll deal with various conspiracies and myths and other stuff.
Long-term continuous overcharging produces an artificially induced drop in capacity that resembles memory. It can also decrease the overall life of the cell. A deep discharge/charge cycle will recover much of the cell's life but long-term damage is very likely. This is not "true" memory because the cell is not subjected to repeated charge/discharge cycles that the cell eventually remembers. It's simply a decrease in capacity due to overcharging, and yes, it is mostly reversible. It is also not memory because the point at which the cell capacity drops out varies with the rate of discharge. The capacity loss due to long-term continuous overcharg- ing is caused by loss of contact of the cadmium hydroxide particles with the negative plate. Electron microscope pictures show that overcharging causes the particles to grow larger, especially at higher temperatures. This reduces the surface contact with the pores of the negative plate. A deep discharge/charge cycle restores the hydroxide particules to their normally smaller size -- increasing surface contact. Overcharging on the negative plate occurs when all the cadmium hydroxide is converted to cadmium metal. Once that occurs, only hydrogen gas and heat are produced (Oxygen gas is produced at the positive plate at the point that it becomes overcharged.) These gases, especially hydrogen, will eventually vent from the cell if overcharging continues, thus reducing the effectiveness of the electrolyte.
The real meaning of memory effect comes from precisely repeated charge/ discharges (without overcharging) of sintered-plate [*] nickel-cadmium cells where the cell seems to remember the point of discharge depth. The effect is exceedingly difficult to reproduce, especially in lower ampere-hour cells. In one particular test program -- especially designed to induce memory -- no effect was found after more than 700 precisely-controlled charge/discharge cycles. In the program, spirally- wound one-ampere-hour cells were used. In a follow-up program, 20-ampere-hour aerospace-type cells were used on a similar test regime. Memory effects showed up after a few hundred cycles. [Test program conducted by Pensabene and Gould at GE, I believe.] This kind of memory appears to be related to the "efficiency" of the positive plate. It seems that repeated precise charge cycles affects the ability of the cell's active chemicals to charge fully, after which the positive plate begins to produce oxygen (as if being overcharged). Hence, it is possible for both gases and uncharged particules to exist simultaneously. Strangely, if the cell is carried out into overcharge the memory effect largely disappears. Hence, overcharging actually reverses the "true" memory effect.
Another reason memory effect is a myth since all the consumer charger's I've seen actually overcharge until there is a slight voltage drop (due to an increase in resistance from the formation of larger cadmium hydroxide particules that cause contact loss). It's because consumer chargers actually overcharge that you have to give the battery a deep discharge from time to time. It has nothing to do with memory.
[*] And just in case you are wondering what a sintered-plate is, the plate is constructed by sintering [welding without melting] a fine nickel powder with a surface area of about one square meter per gram. This produces a honeycombed structure that is about 80% open pores. The negative plate is then impregnated with cadimum hydroxide. The positive plate is impregnated with nickelous hydroxide (which converts to nickelic hydroxide when charged).
And Graham Stoney puts it this way:
The NiCd memory effect business is an urban myth, but it still keeps coming up. In summary, if you overcharge a NiCd battery, it develops a voltage depression which makes the battery appear to go flat earlier than you would expect. Since the discharge curve is so steep, sensitive devices which rely on battery voltage to detect when it is almost flat, will report that it is almost flat early due to the voltage depression, when in fact the cell still has significant charge. The voltage depression can be rectified by discharging the cell to its full discharge level.
Many people misinterpret this phenomemon and conclude that the battery somehow remebers its last discharge level on the next charging cycle. This is not the case. The only effect that the current charge level has on the next charging cycle is that it's much easier to overcharge a NiCd cell whose current charge state is unknown, than it is to overcharge one which is known to be flat.
The so-called "memory effect" is a simple case of user error in overcharging the cell. If you don't ever overcharge a NiCd cell, there's no need to discharge it before recharging it again.
Solution: Quickly post an article like this one - it will quench the flames, educate the original poster, and diminnish any further misinformation and testimonials that usually follow.
A number of articles exist on NiCd batteries which were written by either those involved in the hobby or as a service to its customers by various hobby stores: