I need something with max spead near 400Kbyte/s

Then go parallel. Have a lot of fun getting the '245s synchronized, remember that you can NEVER have 2 that simultaneously output to the bus, not even for a nanosecond. In this respect, do not forget that port pins are high during reset, you may need inverters on the pin that drives the direction.

To obtain 400kbyte, you also need better than 'plain vanilla' '51s since a 12 clocker running 12 MHz can never communicate at 400 kByte.

just see (even assuming the recepient respond immediately, which you can NOT assume)

fetch data  2
inc ptr          2
data on bus      1
set clock        1
test accept      2
clr clock        1
loop             2

A total if 11 clock cycles. So for a 12 clocker you will need 12 * 400k *11 = 52.8 MHZ min clock and the at89c4051 & 80s52 you mention can not run that fast.

Do not forget calculations as above before you put hardware together.

Erik