This scaling computation is too naive and gives the wrong conclusion. If you look at the number of transistors it's about 4.3 million for the BL808 vs 9.5 million for the Katmai core. Counting the logic transistors looks even worse for the Pentium: 2.5M for the BL808 vs 7.5M for the Katmai.

So the Pentium 3 is 2x the transistor count of the C906 (even 3x if you ignore SRAM) and that's despite being 32-bit while the C906 is 64-bit and has a much more capable vector unit.