As a general first note, praestandi looks much more like a gerund than a gerundive here.
A gerundive would be passive in nature.
A gerund is active; it is best understood as a case inflection of the infinitive.
If praestare is "to excel", then praestandi is roughly "of excelling".
For example, ars magna scribendi is "the art of writing great things"; compare this to magna scribere, "to write great things".
One can also use the gerundive (adjective) instead of the gerund (noun) for this: ars magnorum scribendorum.
I see a couple of different ways to read magna in your passage:
Magna modifies the subject: "...so that they, being great, would..." If the subject contains a list of different kinds of things, especially of different genders, a neuter plural is possible. (This is why supplying a semantically understood subject angeli can be misleading.)
This is also not quite this simple nor do I think this reading works here.
I just wanted to point out this direction although it leads to a dead end.
Magna is the object of praestare. This requires preastare to take an object; in many uses this verb goes with a dative, but we must have a meaning with an accusative instead.
Magna modifies agilitate. Then magna agilitate denique summa could be read as "with great and thereafter even greater agility". I'm not sure what to make of the punctuation in the quote, so the lines between different clauses are blurred.
Magna might be slightly misspelled and might modify robore (should be magno) or the subject (should be magni). This is a dangerous route to take, as you are essentially forcing your own interpretation and ignoring the text.
Perhaps magna is indeed the object of valere.
As you mention, the dictionary entry suggests that valere is indeed not used with an accusative object.
I would not be surprised by a pronoun accusative object like hoc valeo for "I can do this"; such usage is possible with many verbs but not with more complicated objects unless the verb is transitive.
In valet tres nummos (mentioned in a comment by d_e) the accusative is adverbial and does not indicate an object.
This is similar to Roma magnum spatium abest, "Rome is far away".
I think number 2 is the most likely one.
The second most likely one is number 3, and the rest strike me as ungrammatical or too stretched.
(My recommended approach is to write down all readings you can possibly think of, not stopping at the first one, and then analyze which ones sound most plausible. Some of the conjectures turn out silly, but you don't always have that retrospective wisdom at square one.)
However you read magna, you have to decide what the genitive praestandi modifies.
I think it modifies robore, so that robur praestandi is "the strength of surpassing".
Thus robur magna praestandi is "the strength of surpassing great things".
(The option with gerundive would be robur magnorum praestandorum, but the author chose a gerund instead.)
Then valere ("to be strong") is done with this kind of strength (in the instrumental ablative).
This leads me to translate (overly literally and more fluently):
…ut robore valeant magna praestandi…
…so that they would be strong with the strength of surpassing great things…
…so that they would have the strength to surpass great things…
You may choose to translate praestare with some other word or otherwise reinterpret words.
My concern here is syntax only, as that seemed to be the bottleneck.
For more on gerunds with objects, see this question about it and this comparison of gerunds and gerundives in the grammar by A&G.
