What's interesting about the "spec war" is that it's almost exclusively Microsoft who's gotten so defensive about it. That should tell you something.
As for the floating-point/integer thing.. I did some more reading, and it turns out that the two of them are largely interchangable. A floating-point architecture can easily handle integer calculations, sometimes even faster and/or more efficiently than an integer-based system.
Here's a link that I found quite informative, since it's less "techy" based than many other floating-point links:
http://lua-users.org/wiki/FloatingPoint Although the link is specific to the LUA programming language, I don't see any reason why the base calculations themselves would change from language to language (it's basically just math, which is universal across all computer systems).
Basically, Allard is either misinformed or he's spewing BS, about the "three times the integer performance", since Cell is perfectly capable of handling integer operations. His misinformation is probably based on the fallacy that Cell has only one "general purpose" processor, while Xenon has three. Hence, "three times the performance". What
he neglected to mention was the SPE's ability to handle integer performance if that's what they're assigned by the PPU. Allard seems to be under the impression that Cell, and specifically the SPE's, can ONLY do floating-point calculations, which is nonsense.