Realistically its still probably going to be a couple years before the full 4.x spec is implemented. Mesa is on a 6 month cycle and at the current pace it'll take at least 4 more cycles or more.
Intel went from "one GL version upgrade" every year to "one" every 6 months, and it seams that in the near future they will try for "two" every 6 months. So i believe they will be there sooner. Also they do experiments with Voxel-Raytracing scenarios and i don't believe that they will stay closed just because tomorrow's synthesizer will be new technology (like some others do). Any way performance is not a problem, as 40core_GT3 with 256bit_FMAC per core is good. The problem is that they use a lot of software rasterizing even to the GPU that seams more like simple CPU and not like ASIC (except TMUs). So they support less GL calls and they crushing Wine support. Then, AMD for other reasons it is not in a better situation.
If only Intel were capable to deliver a graphically powerful CPU, preferably with ECC ram support, I would be sold...
You see, all powerful GPUs have wide dedicated memory bus (128 bits for mid-range, 256-384 bits for hi-end) and fast GDDR5 RAM on this bus. Do you really expect that GPU on relatively slow DDR3 bus (DDR3 is only used on slowest and cheapest graphic cards, better cards come with GDDR5 RAM) which is also shared with CPU itself could show really decent performance?
Keep on dreaming. However it's not like if you can emulate fast dedicated bus and GDDR5 in software :P.