1) in my opinion - yes, as long as you are going to seek for an overall performance maximum, by tuning capital allocation between strategies, based on past data - Vince's OptimalF vector, with 20 elements in your case. An extreme example: a strategy is a slight looser on it's own (big winner from time to time but far too many small losses in between); However, in a portfolio, it might improve the overall performance, having it's big winners just when you need them, when others loose. How much capital to allocate for this looser-alone-but-portfolio-improver? OptimalF is the answer. Using OptimalF still does not exclude decisions like taking out grid a system manually if SMB gives up, you don't have to wait until its OptimalF drops to zero.
3) My favourite books on subject are: R.Vince ISBN 0-471-75768-3 ; Jaeckle, Tomassini ISBN 978-1-905641-79-6 ( mentioned here )