Hi,
Well, all my programs has been written with assembler. Currect project I do, is above 20kB of binary code now (=
As about UART, I cannot understand how it takes only 10-20mks? For example, you have two subroutines:

CALL SUBR_0
CALL SUBR_1

Inside both ones you need to report some debug information via UART. But if you write a byte into SBUF before previous one has been transmitted then it brokes value, yeah? I do not talk about modern 51s which have UART FIFO. Okay, so you need to stay and wait before previous byte has been sent (JNB TI,$ , you know). And it takes some more than 20mks. For example, at 9600 baud it takes ~1ms per byte.
Hmm, some my projects use an interrupt with interval about 50mks (PWM music, for example). So additional 10-20mks kill it with overflood. Usage of 3 LEDs inside ISR is not so fatal (=
Finaly, I do not say that debug-via-UART is bad idea. Sometimes I use it as debug tool, but at another time it is not possibile.
Good days!