So it's 16? I wanted to recommed this and a little less (so 8-16), but didn't remember the exact figure and thought it's 8 (and have no access to bibles. right now.)

But in this case, you don't need to interrupt in this high rate all the time, only while searching for the beginning of the start bit, then you interrupt at the bit rate (after the initial half-bit (sw uarts often do an initial one-and-half bittime to interrupt right at the middle of first databit skipping the startbit, but I like to check the startbit too, for added noise immunity)). And the rapid interrupt may consist of

jb rx_in_progress_flag,rx_routine
jnb P1.0,rx_startbit_begins
reti 

only, which is viable even at 4800*16/sec. But, if the design permits, I would rather use an interrupt to detect startit rather then rapid sampling.
And even if you want to do the majority-of-three-sample thing, you don't need to make an interrupt each 1/16 bittime; you make it 14/16-1/16-1/16, this also is not so bad (and usually noise is not an issue with serial and if it is, the three-samplething will not help, either...)

Jan Waclawek