TalkChess.com

[H.G.Muller]

For me 'equidistributed draft' tested as best replacement scheme. This effectively divides up the hash table in equal sections for each ply. (But in a dynamic way, because you don't know in advance how deep the search will be.)

The point is that deep drafts will save you an enormous amount of work on a single hit, but on the other hand have to be kept around excessively long before you get such a hit. (Because moves close to the root change only slowly, so it takes a long time before you construct the transposition there that will reach it again.) So you might have benefitted more from the table entry by storing 100,000 different shallow drafts in there, giving you 50,000 hits that saved you only a little amount of work.