Those continuous delays really show how difficult it is to design <20nm nodes with old tools, especially 193nm light... seems like painting a miniature with buckets.
Also I'm pretty sure that any apparent problems with clockspeed on these new chips and their release of slower (and smaller) parts first is only due to these low yields and excessive defects.
Of course you wouldn't try to get first a quad core + 48EUs on a new node, keep that for later once problems are solved. Same for any bigger die like six-eight cores (thought actually they might be the same size of a quad + GT3...).
The 65W TDP speeks two things to me: either all the performance upgrades are gone into the IGP, leaving only overclocking CPU wise or finally the tables are more even and both have space to grow: in the same TDP there's a Haswell BGA i7 with 3.2-3.9GHz turbo for the cores, I don't see Intel lowering clocks so at best similar ones or closer to Devil's Canyon.
IGP has increased too and for the same reason of above I'm certain they wont lower performance, still if the rumor of 80% over Haswell GT2 is true there might be almost no increases over old GT3 beside with overclocking.
In the end the choice will be hard between this and a six core Haswell part, unless you -need- IGP and the L4 for... well something.