The latency issues are true, however like any architecture, the CELL requires a specific way of coding for it - were the applications designed to take advantage of it from the ground up, meaning, dedicate specific tasks to their specialized SPU's and make the main core bother with assembling the final product from the received results and that first and foremost, they would have reached peak performance.And no, i do not mean to say Sony was not using state of the art hardware. Quite the opposite, they were using, brand new, underdeveloped, expensive hardware. New and different are not always good, they almost always comes at a cost.
That said, this is not always achievable, which is easily showcased by comparing multi-platform games with dedicated PS3 titles. The Uncharted series pushed the PS3 to the limit and it showed, multiplatform games on the other hand were often buggy as it wasn't very feasable to adjust the engine to this peculiar setup.
Deep down I believe that perhaps the CELL would be "better" in a non-gaming device - it's one of those CPU's that's great at doing "a whole lot of stuff at once", and I see it in science-related equipment rather than consoles. It also has quite a remarkable capacity for floating point - of course not as great as the average GPU or Tesla cards, it's more like a bridge in-between a Integer-oriented and a Float-oriented processors, making it perfect for "experimenting".
For the PS3, your average PowerPC build would have been a better choice, and I won't argue againts that, but I wouldn't say that the CELL was tragic - its problem was derrived from how differently it worked from a standard build rather than from some inheritent limitation.
...but that's all just techie banter unrelated to the topic at hand.