"However, the performance increase of Athlon 64 processor in 64-bit mode is a bit higher and equals 35%, while Pentium 4 managed to benefit by only 25% of extra performance."
These are not correct % numbers!
A64 does 6734 msec in 32 bit and 4312 msec in 64 bit. This means it is 56.2 % faster in 64 bit mode!
P4 does 11094 msec in 32 bit and 8203 msec in 64 bit, that means it is 35.2 % faster in 64 bit mode.
"In this case the performance increase by the CPU with AMD64 technology made 55%, while in case of a CPU with Intel EM64T technology the performance in 64-bit mode grew up by 40%."
Again, completely wrong numbers:
A64 does 4.34 sec in 32 bit and 9.61 sec in 64 bit. This means it is 121 % faster in 64 bit mode!
P4 does 5.33 sec in 32 bit and 8.97 sec in 64 bit, that means it is 68 % faster in 64 bit mode.
These kind of math errors can easily be avoided by approximizations. Think of it like this for Mini-GZIP A64 case:
"If it takes about half the time to perform a task, how much faster is the CPU? It's twice as fast, right. So twice as fast = 100 % faster. Why do I write 55 % faster in my article? Because I turned the math around to incorrect numbers. I better check my math..."