But what about P3.3? According to page 18 you need a 4µs high pulse on that pin before you can verify data - according to page 10 it should be low. If my experience is correct page 10 has the correct version...

Yes, there is difference in the table (page 10) and waveform (page 18). You should ask Winbond people about that (and hopefully post here what they say).

One thing that isn't mentioned anywhere (I think) is the length of the CE' pulse for erase. I used 100ms (from the EPROM datasheet), and that seems to work fine.

Yes, it is not sepcified.

I feel that they have made it too complex to program the chip. May be they have done it for gang programming purposes.

PS: I am only talking about W78E52B