PCON.7 or SMOD only affects Timer 1. Using Timer 1 with SMOD=1 and TH1=FEh will yield a baud rate of 9301 with a 3.1% error.
If you absolutely must use this clock freq. and the baud generated won't work with your device then you may need to consider using an external UART with its own clock crystal.
In most 8051 architectures the machine cycle is usually 12 clock cycles. The number of machine cycles required to execute instructions is one or two except for mult & div which take four. There are derivative chips such as the Dallas DS89C420 where the machine cycle and clock cycle are the same. Whatever the architecture, the crystal (or clock) input determines the speed that the MCU operates at.

Hal