>>second Naomi's vector of attack, if the fancy stuff allows it.
>>A 2 step might still be faster, if 1 step is not doable:
>>Alter with defaults only, then update as update logic is often better documented then
>>alter fringe parts and should still be faster for anything involving more than 3-4 alter lines previously.
>
>Do you mean you support that approach or we have to be careful using it?

I support the approach as the least cost incurring from a machine POV.
Not having any idea on the nature of the fancy stuff makes me cautious ;-)

I am pretty certain testing of value-alteration is more intensive with update,
so if it gets really fancy, you might run into real trouble, like bugs in the product,
or short-time trouble, having to test and seek out docs on fringe cases seldom encountered in such combination.

As you are not a newbie and had the same first optim thoughts, I guess you already ran into something unexpected.
A good compromize between your efforts and runtime might be to use the 2-step approach I spoke of...