Hi Eric,

Absolutely correct. I finally wrote the pseudo-assembly code and did the calculations. (sorry for my being 3Л/2 late!) Uploading the data to the paralell latches while having the display off in order to prevent the "shadow"s(ghoasting) consumes some 300us (11.05) which was almost all the time I had in the 32x100Hz-invoked timer interrupt routine. Had I done that, I wouldn't have got nothing but some 1/30th of 1/32 ~= 1/1000 duty cycle. The serial approach is not just a well-engineerred approach, but a must. You saved my gluteous maximus!
The new timer ISR responsible for refresh task and HC595 upload is this:
timer0_isr:
inc 5x32 value(a static value)
if (5x32 value==32) 5x32 value=0
mov dptr,ram_addr.(a global value)
loop:
movx a,@dptr
inc dptr
mov Pz,a ;serial data
setb Px.y ;serial clock
clr Px.y
goto loop 32 times
off ;turn off the display
mov py,5x32 value ;put new value (enabling next column)
setb px.w ;strobe
clr px.w
on ;turn on the display
reti

Best.