I think many people commenting here are looking at the term "APU" too optimistically. IMO, "APU's" were implemented to save platform costs by bringing the integrated graphics on-die (see SoC) first and foremost. As a side effect: Until that point, if you were running off on-board graphics, it was an anemic chip, oftentimes with a tiny little passive heatsink or none at all. Integrating the graphics onto the CPU gives it a bigger (usually active) heatsink (more thermal headroom = more performance) and puts it closer to the CPU.
Also, Intel has had "APU's" since...Clarkdale in 2010? AMD just coined the term "APU" for their Llano chips (2011) with on-die graphics. Since graphics was brought on-die, there have obviously been evolutionary steps that have improved IGP performance (process node shrinks, architecture changes, etc). AMD saw that they couldn't beat Intel at CPU performance, so they dedicated more effort/die space to the GPU. Intel has (roughly) kept pace with AMDs IGP improvements while focusing on power efficiency.
That evolution has led us to where we are today as IGPs can actually begin to reasonably replace dGPU's for low-level 3D gaming needs (not to mention you no longer need a dGPU for Office-use computers, HTPCs, and the majority of consumer content). However, physically, a dGPU will be able to perform better than an IGP simply because of thermals. dGPUs have their own separate cooling solution as well as a boat-load of dedicated power input. The modularity of dGPUs is obviously a bonus as well. Although IGPs will continue to improve, it's not economically worthwhile to give every PC user the equivalent of a 7970 (for example) in IGP. It costs money to build, and people won't see the need to pay that much money for, say, Office-type PCs that will never use that much power. The level of IGP performance will always increase, but the top end of the graphics market will be occupied by dGPUs for the foreseeable future.
Main: i5-3570K, ASRock Z77 Pro4-M, MSI RX480 8G, 500GB Crucial BX100, 2 TB Samsung EcoGreen F4, 16GB 1600MHz G.Skill @1.25V, EVGA 550-G2, Silverstone PS07B
HTPC: A8-5600K, MSI FM2-A75IA-E53, 4TB Seagate SSHD, 8GB 1866MHz G.Skill, Crosley D-25 Case Mod