Taking note that a cheap xtal is about 50 parts per million, so anything that gives 22.579,200 +/- 1,128.96 Hz (22,578,071 Hz to 22,580,328 Hz) should be 'close enough' to spec that nobody can tell (it makes the difference of 0.2s after playing audio for an hour...).
Options
a ) Use the clocking wizard to generate a PLL and end up with 22.580,645 MHz (0.23s of error after an hour)
b ) Use a PLL Primative, with multiply 7, divide 31 and end up with 22.580,645 MHz (once again, 0.23s of error after an hour)
c ) Use one PLL set to multiply 9, divide 35 to give 25.714,285MHz, and feed that through a second PLL at multply 36, divide 41 to get 22.578,397 MHz (0.13s out after an hour)
d ) Use three PLLs, multiply 8 divide 25, multiply 21 divide 25, multiply 21 divide 25 to get 22.579,200 MHz - bang on (however I don't think that Nexys3 has enough clock management tiles for this...)
e ) Exotic stuff - perhaps use a polyphase filter to resample the 448100 sample stream to different sample rate, perhaps 25MHz / 512 = 48.828,125 samples per second, then use a 25MHz I2S clock, which should at worst add 1 least significant bit of noise - but why would you bother?
f ) Add an external clock source that is a more friendly frequency.
If I ever need to do anything tricky with clocking, I find it better to use clock management primitives than to use the Clocking Wizard, as you get exactly what you ask for. Once you've done it once it is a lot quicker too!