Perhaps you could explain how your current logic works for that test (attempting to read the code and it isn't making sense yet...)

From my understanding:
With the NES (NTSC, PAL) and Famicom, reading the palette is impossible; attempting to do so will simply return mirrored nametable data (goroh has verified this on a Famicom, and Kevin Horton has verified this on a CopyNES).

With the Dr PC Jr, it is possible to read the palette data, and there is no 'delay' when reading said data (as there is when reading from hardware external to the PPU). This is most likely because the people who RE'd the PPU were lazy when reimplementing it (not making /W a *requirement* for the internal palette bypass circuitry).

--
Quietust
P.S. If you don't get this note, let me know and I'll write you another.