"SDP" is just TDP when running tablet workloads. Exynos Dual will consume ~8W in power-virus like tests if the power management unit does not throttle down.
As a general rule, 5W TDP is the target for a fanless tablet. Of course 6 or 7 or 8W can be made fanless as well depending on the cooling but this would be on the high side and increases the manufacturing costs for better cooling. There is a good reason why Bay Trail-T is around 4W TDP. Haswell-Y with 6W SDP has a 11.5W TDP in nominal mode and 9.5W in TDP-down/LFM mode. For a tablet this is still a very high power consumption
You misunderstand Intel's TDP and SDP targets. Only one of these will be used for any given design. You won't have a tablet that simultaneously dissipates TDP and SDP. If an OEM designs to SDP, they will load a value into the PL1 register of the CPU that's consistent with this power, and the CPU will regulate operation such that it ensures the steady-state cap on power.
So if an OEM designs to a 6W SDP, then the 11.5W TDP will be ignored by the CPU. Of course, that's just for steady-state operation. Short bursts will still be possible, to an arbitrarily high power state. This is how "turbo" works. In Ivy Bridge, that number is set to 21.25W, which is the 17W TDP plus 25%. They kept the same turbo number for the Y products, meaning that the 7W SDP is the steady-state power cap, if an OEM chooses it, while at the same time they can run at 21.25W for brief periods.
Brief periods are usually 1-30 seconds, as defined by another register. Just read the specs - it's all there. The idea is that even a constrained tablet can run in turbo for a number of seconds (or even 10s of seconds) before reaching critical temperatures, after which it will return to the steady-state power cap - whether that cap is defined at the SDP, or the TDP (the OEM will choose only one).
So to answer your question, the 6W Haswell has a range of possible power caps, from 6W to 11.5W, and the 4.5W Haswell extends the bottom of the range to 4.5W, while keeping the top of the range. This gives OEMs a ton of flexibility, depending on the cooling requirements of their designs.
Setting it to TDP will mean that all workloads will have their minimum clocks reaching the stated frequencies, even while hitting steady-state workload conditions. Setting it to SDP will mean that increasing amounts of throttling will be necessary, up to and including reaching the LFM frequencies. However, since different workloads have different activity factors, you won't necessarily see all of them running at LFM under steady-state. In fact, many workloads will still run at fairly high frequencies, especially single threaded workloads, since one CPU core with one thread can clock higher at a given power cap than 2 cores with 4 threads.