I just don't believe that about mechanical micro switches being the most accurate. Here's my take, coming from an electronics background;
1. Optical interuptor sensors. Lousy accuracy, affected by ambient light, IR, even radiated IR (heat). The optical receiver is not temperature compensated and worst of all they age over time so the receiver properties change over time. Also a big problem with dust of any type.
2. Mechanical microswitches. Even if this was to have the initial repeatability of a "no moving parts" hall system, there are parts that will wear over time and change the home point. They usually incorporate a plastic lever and pivot, and plastic wears. It also has a high thermal expansion (much greater than metal) so if the inbuilt lever is 5mm in size and expands 2% over normal operating temps (quite typical) that would be an error of 0.1mm even before the wear starts to occur. Contact wear is also a given, and any mechanical interface may get some dust or swarf between the lever and actuating surface at any time.
Most of these error problems also apply to hall sensors that have a mechanical lever interface (like the type mentioned by CreviceReamer above) which is why I would avoid these type of hall sensor.
3. No moving parts hall switch. Using a "slide by" operation gives very high repeatability, probably as good or better than 0.01mm. There are no parts to snag or trap swarf. Obviously these are not suitable for cutting steel or ferrous metals as the magnet will attract steel swarf but wood, plastics, aluminium, brass etc are ok.
The hall switch has 4 sensors in a bridge so it self-cancels ageing effects of the silicon. Also some brands are temperature compensated for both the hall switch AND the magnet itself (see the datasheet) so even though temperature changes may affect the silicon and the magnet strength that is all compensated out.
When you crunch the numbers, the hall switches look miles ahead for accuracy and repeatability! I'm really looking forward to seeing Vladimir's (vsengineering) test results as his machine allows much finer testing than my machine results shown on page 1 of this thread (which are only in resolution of 0.01mm).