Your information is highly dated and perhaps your sources are also a bit biased. At any rate, 100x performance hit is stupid wrong.
Static translation was achieving 50-70% native performance rates (measured against clock cycles) with FX!32 on Windows NT for Alpha in the mid 90's. The problem of course has been very well studied since then particularly with the advent of virtualization and the x64 instruction set and the need to enhance the performance of x86 code running on even Intel's own platforms. Furthermore for any particularly glaring issues that are the fault of the hardware -- well it is much more easily tuned today than it once was. A bespoke opcode or extra register to assist in a specific task is no longer a monumental engineering undertaking today -- it is a matter mostly of dev/test/validate.
The approach taken with x64 to support x86 native execution is quite different than attacking the problem with emulation. Is there a performance hit? Certainly, but a hit of 10-20% simply doesnt make up for the fact that you might be able to have an 8 core ARM for the same price and power budget as a 2 core x86 mobile cpu. The applications that lose in this scenario are the ones the rely on raw single thread performance. Certainly some games are in this camp, but many games which make efficient use of threads are not.