Right. 7135 drivers burn up excess voltage as waste heat. 7135 driver efficiency is basically Vload/Vbatt. So on the one hand, you would like to match Vbatt as closely to Vload as possible to get the max efficiency and least amount of waste heat generation. That works great if your power source is a bench supply or something like that. It gets trickier when the power source is a battery pack. The challenges are that you don't generally have infinitely selectable voltage, unlike with a bench supply, and the voltage out is not constant with batteries. With batteries, your voltage levels are quantized in steps of say 1.2V for NiMH or 3.7V for Li-ions. And the voltage out continually changes under load. So you might start out at 1.5V with NiMH or alkalines, immediately sag to some lower voltage when under load, and keep drooping from there (at different slopes depending on cell chemistry, cell quality, cell size, current draw, etc).
Thus, not only can it be tough to match Vload with Vbatt, but Vbatt doesn't stay constant.
You also probably want to have some voltage headroom so that you will have decent runtime in full regulation. You don't want to have Vbatt exactly equal to Vload. You need the headroom not only because of the non-constant battery voltage, but also because 7135 drivers require 0.12V of headroom and the flashlight system will also have parasitic resistance from things like battery contacts, spring contacts, and switch contacts.
When you have a low Vf LED, that can make it easier to power with a single Li-ion or perhaps 3xNiMH (or maybe 3xalkaline) if the current draw isn't too high. Something like an XM-L *might* work, for example if you ran it at 1A forward current. The XM-L's Vf is probably 2.9V or so and 3xD NiMH and 3xD alkaline both can stay above 3.6V for a fairly long run time. But if you try to power the same XM-L at say 2.8A drive current, now Vf could be 3.35V. Add the 0.12V for the 7135 overhead and you need a minimum Vbatt of about 3.5V. Since the current is also greater, parasitic resistance becomes more important in terms of parasitic voltage loss. An 0.1 ohm total parasitic resistance means a voltage loss of 0.28V, meaning you now need about 3.75V min Vbatt. 3xD NiMH can still deliver that, but for how long relative to the regulated run time that you desire? I doubt that 3xD alkaline will ever run in regulation. A big 1xLi-ion setup could hold above 3.75V for a while, but again, how long relative to your requirement and relative to the total capacity of the cell?
So 3xNiMH and 1xLi-ion give you best efficiency, but staying in regulation can be tricky depending on a host of interdependent factors.
But the straightforward solution of adding more Vbatt costs you in terms of efficiency/waste heat. You definitely can go with 4xNiMH and stay comfortably in full regulation. But your driver efficiency will probably start out around 70%, which is a bit low. Personally, I would not worry too much about that. 7135 drivers are rugged and you can always heat sink the driver. It gets tougher if you start piggybacking a lot of 7135 chips to get much higher drive currents.
However, you can't go with 2xLi-ion, since Vbatt will exceed the maximum rated input voltage of the 7135 chip. You risk frying the driver.
My 7135 setups generally use modest drive currents and low Vf LEDs such that 1xLi-ion delivers more than enough voltage. If I need more drive current, which increases Vf, I typically go for buck drivers (or boost drivers).