A few observations...
(a) re - electing to operate to the requirements for a higher category - there ought not to be any concern with this as categorisation is min speed dependent and operating to more restrictive standards should be viewed favourably by company lawyers, the Regulator, and the judge in the enquiry/action after the mishap ...
There is never any requirement which says that one MUST operate at the most critical limit - it may be a useful option, but not a prescription. Thus, to descend to the published minimum, for example by day in visual contact after a cloud break, if the circuit can be flown at a higher, more "normal" height, seems a little silly from a risk management viewpoint. This would be philosophically similar to conducting all turning operations at the limit load factor.
(b) if you elect to operate to a higher category, then it has to be all or nothing - while you get the benefit of an increased radius for circling protection there comes a downside with the attendent larger protected area - you also probably pick up some additional hard bits and this leads to the typically associated higher circling minimum altitude.
(c) be VERY, VERY wary about descending below the published minimum until on the normal approach path UNLESS you know VERY ACCURATELY what and where the obstructions are. It is very illustrative to check with the local procedures designers as to how they establish the minimum.
Unless someone actually goes out and does a survey of critical obstructions, and this is done in some cases by individual operators seeking either an alternative or just a better documented procedure, one ought to be very careful and cautious about the accuracy of the obstacle data used in routine procedure design.
The flight inspection test of the procedure (presuming that is, in fact, done) will pick up any gross errors but won't necessarily cover the brave chap stooging around at some significant height below the published minimum. It is all well and good to bandy intentions and expectations embodied in PANS-OPS or similar diagrams. However, at the end of the day, the operator or pilot who wants to predicate his actual height on the specific flightpath needs to be very sure of the specific obstacle profile.
(d) the question of a missed approach initiated from a circling approach is always contentious.
The simplest philosophy says that one turns toward the field and then tries to pick up the letdown published missed approach. This is often not so simple a matter.
I suggest that a better management procedure is that an operator, choosing to conduct (or permit) circling approaches at a specific aerodrome with other than benign surrounding terrain, ought to conduct an escape analysis similar to that which should be done for the takeoff case. After all, the two are much the same sort of thing, both from the point of problems and desired outcome.