Maybe its just the way I wrote my code, however I dont see any other
    way to do it.  I suppose I could take out my delay's, but still it
    would take me around 500us to read + - 100us but not 20us.  Im using a
    24LS32 serial eeprom from Microchip.

Hmm.  I should preface this by saying that I haven't actually DONE
this, but...

The way I read the 93c46 datasheet, reading a byte of data consists of
clocking in a 9bit read command (which includes the address), and then
clocking out 9 bits of the data (start bit plus 8 data bits, essentially.)
The clock can be up to 2MHz, so in theory you could do the byte read in 18
* 1/(2e6) or about 9 microseconds.  Now a 4MHz PIC certainly can't generate
a 2MHz clock signal in software - I'd guess that it takes about 4 to 5
instructions to do each shift & clock, perhaps more.  Call it 8
instructions per bit to come up with an even number (2us), and your random
read should take about 36 us.

Now, on the 24c32 parts, it looks like a "random read" is about the slowest
possible operation you can do, (write device address, "write" comand, word
address, device address again (aborts the "write"), clock out data...)  On
top of that, the clock spec if 5x slower than the 93c46 (400kHz vs 2MHz),
so it doesn't sound to me like your code is badly written, it just sounds
like you picked the wrong eeprom...  (OTOH, the 24c32 part has a trivial
"read the next sequential byte" that could be utilized.)

I had noticed that there were different varieties of EEPROMs of varying
complexity, but this is the first time I've noticed the substantial speed
differences!  I think I'll have to pay more attention.  The 93C familly is
fairly low density, but the SPI EEPROM chips (AT25* for atmel) have
densities in the same range as the 24C parts, and behave similarly to the
93C parts... (don't seem to be as available, though.)

BillW