I must say I have enjoyed reading this thread immensely. Thanks to all involved. I decided to put in my tuppence worth, so here we go.
An Epicurean pleasure is one of the canonical faculties which makes it completely independent from reason. Canonical pleasure is a feeling but "feeling pleasure" is something completely different and these two are not interchangeable. What we commonly describe as our feelings (pleasure, desire, grief, sadness, joy etc.) is our awareness (= reasoning) of a situation we find ourselves in. In this sense, our feelings have nothing to do (as they are dependent on our reasoning) with canonical feelings in Epicurean terms.
Desire is a non-canonical feeling. It's our awareness (again, cerebral activity) which signals to us that we find ourselves in a situation that is not optimally pleasant for us and we crave the change. "Feeling desire" is always linked with "feeling pain". Similarly, "feeling happiness" is always linked with "feeling pleasure".
The problem seems to arise when we mix canonical pleasure with "feeling pleasure" (= mental awareness of being at the right time in the right sandbox with the right toys) and try to decide how this "Frankenstein pleasure" measures in relation to desire. Not a good idea. To sum up, I'm thinking this:
1) desire is always painful when we treat desire and "feeling pain" as conscious feelings (awareness of our situation)
2) Epicurean canonical feelings of pleasure and pain cannot equate to any conscious feelings (including desire)
Epicurus' classification of desires shows really well that Epicurus knew that desires are feelings in the sense of conscious (mis)understanding of our situation. He knew that when reasoning is involved people are prone to make mistakes in their judgements. Types of desires is a neat tool to increase our chances to reason well and to make choices that minimise pain and maximise pleasure.