I received six off-list responses, alongside three onlist ones.

LIKES

Simple to get running, a lot of bang for the buck (3 mentions)
Checksum checker (3 mentions)
HTML display engine (3 mentions)
Search engine: "simple and fast" (2 mentions)
Manakin theming (2 mentions)
Legible displays, easy structuring of data and depositors into communities/collections
Flexible Dublin Core, easy defining and rebuilding of indexes
Storage in SQL database
Easy to showcase and share data
Community spirit and visionary dedication!
Configurable submission workflow, metadata templating
Maven dealing with Java dependencies

NEEDS/ISSUES

Complexity of authorization/permissions system, poor fit with real-world workflows, too much work for DSpace admins that can't be delegated (5 mentions)
Communities/collections model confusing and unintuitive for end-users (3 mentions)
Submission process needs streamlining and simplification; input-forms.xml needs to be end-user editable (3 mentions)
Allow bitstream updating/addition after deposit by users and collection administrators (me, George)
Doesn't automatically feed people into user groups based on LDAP group membership
Allow depositors to withdraw their own items
Difficult to customize; also, moving to Manakin costs functionality
Documentation scattered and confusing
Can only use Postgres and Oracle databases
Too-close integration with handles

I tried to exclude this from the question, but I got a lot of it anyway!

better i18n (community/collection descriptions, metadata-input forms) (3 mentions)
statistics (per item, per author) (3 mentions)
batch import of citation-only references from a single document
web UI to prompt a reindexing
add "also by" (author) or "see also" (subject) links to item pages
persistent bitstream handles/URLs
HTML in metadata (e.g. abstracts)
embargo support
bulk metadata editing through a web UI