If your timer is interrupting 4800-times per second, then you take samples from a random part of the serial stream's pulses - once too close to the leading edges, then too close to the trailing edges. A small asymmetry in pulses' shape, and/or edge jitter and you have your 50% error rate...

You need somehow start your sampling around the middle of the edges. Traditionally, sw uarts' RxD are connected to p3.2/p3.3, and exteral interrupr is used to detect the start bit's (1->0) edge, then disabling the external interrupt, start timer, sample the start bit after a half-bit's time (in the middle of bit), then take the databits' and stopbit samples in a bit's time.

If you insist on p1.0, you need to take your samples more often, at least 4-8x faster to detect the start bit edge more reliably. Then proceed as described above.

On a 52, maybe some trick with timer2 can be found, but I am not sure...

Jn Waclawek
Jan