I'm quite surprised that it only took you about half an hour to check all ED puzzles of F2. Did you find some further reductions in the number of ED puzzles?
Thank you for your confirmation of F1 pattern's invalidity!
Methods I used for patterns F2-F5 varies. All searches were performed on my notebook (single core was used) - HP Compaq 6710b (Intel Core 2 Duo CPU @ 2.4 GHz).
F2 pattern was searched through exactly by blue
's method, described in details at setbb.com forum in 2010 (concerning exhaustive search through Fractal Pattern). Key points of this method (in my interpretation).
1. All possible completions
(not puzzles) are checked - do they contain unhitted UA sets?
2. First B345 band and B258 stack are filled and checked for unhitted UA sets. If they have no unhitted UA sets, we fill 4 corner boxes (B1,B3,B7,B9). When all corner boxes are filled, we impose on them all possible clue pattern templates (81 templates). Those corner templates are produced from given pattern (F2 pattern) by all permutations of r1,r2,r3,r7,r8,r9 rows and c1,c2,c3,c7,c8,c9 columns.
3. JCZSolver is used in checking completions for unhitted UA sets i 2 steps. First, completion "converted" to puzzle by solving completion by solver, second - the puzzles was solved. If the puzzle has only 1 solution, completion doesn't contain unhitted UA sets.
It takes me 162 seconds to find all 18432 crossings (band+stack), containing B5 box (exactly the same as for Fractal Pattern) and having no unhitted UA sets. Checking those 18432 crossing with corner boxes takes 49 minutes (I was not accurate saying "30 minutes").
Unfortunately this method don't work for F3 pattern. It has too many crossing, containing B5 box and having no unhitted UA sets. So I was forced to change algorithm (it took me a month or more of hard work). The best result was achived when first all crossings w/o unhitted UA sets, containing B2 box (B123 band + B258 stack) were found, then the rest 4 boxes (B4,B6,B7,B9) were checked. Somehow It was faster to change clue pattern templates (81) for those boxes first, then to fill free cells to get completions, check them for unhitted UA sets, etc. Search was done in 88 hours.
Absolutely the same method was used in searching through F4 pattern. I have no time to rewrite isomorphisms checking procedure and comment out the code specific for F3 pattern. Nevertheless, search through F4 pattern was performed in 53 minutes! It was a pleasant surprise for me.
F5 pattern was processed in the same manner as F4 pattern (without rewriting isomorphisms checking procedure), but I was forced to change boxes processing order. First crossings, containing B1 box (B123 band + B147 stack), were processed, then the rest boxes. The search was done for 13 hours.