One reason you didn't see many Z80B machines is that not much was available in the way of affordable support chips for it. For example, look at the advert you cite: it has prices for PIO, CTC, DMA, and SIO chips to work with either Z80 or Z80A, but for the Z80B there are no DMA or SIO chips listed, and the PIO and CTC chips are much more expensive. You could also use a Z80A with one of Intel's DMA chips, but those only went up to 5MHz, so if you needed DMA, you were only looking at a 25% increase in performance at most, not 50%.
Memory was a problem, too: for a Z80B at 6MHz to run with no wait states, it would need memory with a cycle time of 1.5x the clock rate, which unless you were using page mode (and consequently more expensive support chips) meant you'd need fast RAM chips too (a 4164-120 should do the job, but they were quite a bit more money than the 4164-150s that most computers were using at the time). And you'd still need logic to generate the RAS and CAS lines, which needs to be able to run faster than the CPU clock in order to produce the signals correctly. For example, the Sinclair Spectrum generates these using its ULA, which runs internally at 4 times the CPU's clock rate. So to make a Spectrum-like computer run a Z80B at full speed would need its ULA to run at 24MHz, which was pushing the capabilities of programmable logic at the time, so would likely have required a full custom ASIC, pushing it beyond the capabilities of many manufacturers. It could have been done with TTL logic chips, but would have made the resulting boards quite large, and expensive to produce.
Now, there were manufacturers who had the capabilities to do this, but in reality they had no real desire to: Commodore could have done it, but were completely invested in the 6502 series of CPUs due to their ownership of MOS Technologies. Acorn could, but then by the time the need for a faster mainstream microcomputer became apparent, Acorn were already thinking along the lines that would lead to the ARM processor.