diff --git a/cds_dojson/matcher.py b/cds_dojson/matcher.py
index c6adf76..ae6f15a 100644
--- a/cds_dojson/matcher.py
+++ b/cds_dojson/matcher.py
@@ -28,6 +28,12 @@
 from invenio_query_parser.walkers.match_unit import MatchUnit
 from invenio_query_parser.walkers.pypeg_to_ast import PypegConverter
 
+# Cache of (name, model, parsed_query_ast) tuples keyed by entry_point_group.
+# Entry points and their query ASTs are static for the lifetime of a process,
+# so scanning importlib_metadata and re-parsing pypeg2 queries on every call
+# was the dominant cost (~84s for 1320 calls).
+_models_cache = {}
+
 
 class Query(object):
     """Query object."""
@@ -35,18 +41,33 @@ class Query(object):
     def __init__(self, query):
         """Init."""
         self._query = query
+        # Parse once at construction time; re-parsing on every match() call
+        # via a @property was the other half of the cost.
+        tree = pypeg2.parse(query, parser, whitespace="")
+        self._parsed = tree.accept(PypegConverter())
 
     @property
     def query(self):
-        """Parse query string using given grammar."""
-        tree = pypeg2.parse(self._query, parser, whitespace="")
-        return tree.accept(PypegConverter())
+        """Return the pre-parsed query AST."""
+        return self._parsed
 
     def match(self, record, user_info=None):
         """Return True if record match the query."""
         return self.query.accept(MatchUnit(record))
 
 
+def _load_models(entry_point_group):
+    """Load and cache entry point models for a given group."""
+    if entry_point_group not in _models_cache:
+        entrypoints = set(importlib_metadata.entry_points(group=entry_point_group))
+        models = []
+        for ep in entrypoints:
+            model = ep.load()
+            models.append((ep.name, model, Query(model.__query__)))
+        _models_cache[entry_point_group] = models
+    return _models_cache[entry_point_group]
+
+
 def matcher(record, entry_point_group):
     """Matcher for DoJSON models.
 
@@ -60,15 +81,12 @@ def matcher(record, entry_point_group):
     logger = logging.getLogger(__name__ + ".dojson_matcher")
 
     _matches = []
-    entrypoints = set(importlib_metadata.entry_points(group=entry_point_group))
-    for entry_point in entrypoints:
-        model = entry_point.load()
-        query = Query(model.__query__)
+    for name, model, query in _load_models(entry_point_group):
         if query.match(record):
             logger.info("Model `{0}` found matching the query {1}.".format(
-                entry_point.name, model
+                name, model
             ))
-            _matches.append([entry_point.name, model])
+            _matches.append([name, model])
     try:
         if len(_matches) > 1:
             logger.error(