The ISP algorithm from Winbond works well. They have 2 PROM areas - the conventional 'APROM' (64k in this case) and a 4k Boot LDROM area. The latter is overlaid and invoked by either holding down P2.6 and 2.7 at reset, or by switching to it via code written by the programmer in the conventional APROM. Execution commences at 000h in the LDROM after either methods of entry - the interrupt vectors are at the usual locations, but in the LDROM of course. PROBLEM is that both PROM areas come from the factory cleared, so not only do you have to write the loader software and, if you need it, loader invoking routine within your application, but you have to load them initially with a parallel programmer. I think these are not common for the Winbond - we had to have bespoke software for our ALL-07 programmer in order to do it.
The code from the data sheet should work, 'out of the box'. I have modified it for special applications.
Dave