A corpus search shows 18 hits for scibam including four with ne- and 4 hits for scibat with none with ne-.
The variants sciebam and sciebat have 79 and 145 hits (15 and 22 with ne-), respectively.
Overall, it seems that the imperfect endings without the -e- are rare.
For example, I found only one occurrence of audibat but 36 for audiebat.
Based on this, I would say that nescibat is a valid variant but just happens not to be attested.
Dropping the -e- is possible but so rare that it does not occur for all words in all forms in the attested literature.
I expect different authors to have different spelling preferences, but I have not looked into this aspect further, as it is beyond the scope of your question.
And it is certainly possible that different editors have their own preferences too, but based on the numbers I found, I see no reason to argue nescibat to be invalid.
I think declaring nescibat invalid is reading too much into the attestations.
However, it should be noted — as C. M. Weimer pointed out in a comment — that most if not all of the attestations appear to be from poetry.
Therefore the transformation sciebat > scibat may be taken as metric flexibility rather than a feature of normal speech.
But being as common as the phenomenon is, I find it hard to believe that it would sound very alien to a Roman ear.
A more careful formulation of the answer is that nescibat is certainly valid in poetry, but its status is less clear in prose.