Well, one thing I don't know if whether thats a full vector op and scalar (i.e. 4 component ops and a scalar) or not. Onther element to consider is whether all the ALU's in the pipeline have the same capabilities as one another, which given the unified method one would assume they have to be - this is not something we've seen in the desktop space yet since although we have 32 ALU's they have not all been of the same capablities.
So, on the whole I think 48 ALU's should be fairly powerful. As I mentioned previously, the biggest element of this is not necessarily the quantity of ALU's, but the scheduler and how well it can opporunistically fill instruction slots and keep pipeline bubbles to a minimum - get it right and you can end up in greater efficiencies than we've seen in traditional pipelines, get it wrond and it could be much worse. IMO this is probably the area that the most work has been going in to since the original R400.