Since hardware with full support for floating point textures and frame buffers is so ubiquitous today, is there a reason to shoehorn in HDR using non-float formats? The doubling in memory usage incurred by half-floats is indeed daunting, and minifloats just don't offer the desired level of precision in many cases. Still, this could mean so little on contemporary hardware that the benefits far outweigh any potential penalties, 

I haven't bothered with manual HDR encodings in many years. If you really want to fit into 32 bits per pixel, then the R11G11B10_FLOAT is nearly ubiquitous on PC's (it's a required format for DX10-class GPU's, which have been around for a very long time). The only valid use case for the old encodings might be on older mobile hardware, but that's outside of my realm of expertise. 

In the AAA space a lot of post processing / lighting etc has a ton of man hours poured into optimisation, including finding which bits of work can be done in the same shaders, instead of multiple passes. This pass-reduction kind of work can reduce the bandwidth impact of "fat" HDR formats. So - if your renderer isn't as hyper optimised, you'll likely see a larger difference between thin/fat HDR formats!

Fp16 is also nice in that it's filterable, whereas manually packed formats (logL, RGBM, etc) need to be point sampled, manually decoded, and manually filtered (if required). This can make effects such as blurring quite expensive (perhaps a worse cost than the bandwidth cost of going "fat"). 

Fp11 does have quite a limited range - technically it has the same range as Fp16 (around 65k), but because it doesn't have many mantissa bits it suffers from a lot more color banding. To get similar color banding quality to sRGB 8, you need to add a manual gamma curve on top of it with an exponent of between 2 to 3, which gives you a maximum value of between around 255 (gamma 2) to 40 (gamma 3). That's still enough range for HDR if you apply camera exposure in your lighting shader, but not enough range for general purpose use if you apply camera exposure in your tonemapping post process pass.