The decision to use eth or thorn in a given circumstance seemed to have been largely personal preference; something that becomes clear when reading OE pieces by different authors! I think there were some general tendencies, but it doesn't seem like there were many absolutes. Given that OE spelling was far less standardised than that in modern English, as Peter observes, this would have been true, to varying extents, of many words with spelling variants. I have certainly found this to be true when reading long passages of OE even by one author, who will flip between different spellings from page to page. Although I think it's always important to remember that in amongst all this fluidity, spelling was far more consistent in OE than an "anything goes" approach.
Launching into a bit of speculation, although modern OE scholars have noted obvious broad dialectical (and temporal) variants, the learned individuals who produced OE literature must surely have been familiar with variants outside their own time and place, and perhaps used those they liked to add a bit of spice to their writing - particularly as OE prose is notably prone to repetition and recapitulation! - which might result in what we perceive as inconsistencies in our post hoc classifications? Also, much OE writing survives as copies produced by later medieval scribes, who may have consciously or unconsciously substituted letters here and there, based on their own preferences and background? There's a brief section related to this in David Crystal's excellent “The Stories of English”, where he examines variations between multiple copies of the same manuscript, although I think they were all AS in date.
