diff --git a/jedi/inference/references.py b/jedi/inference/references.py
index f9263315..e5115ca1 100644
--- a/jedi/inference/references.py
+++ b/jedi/inference/references.py
@@ -11,8 +11,18 @@ from jedi.inference.gradual.conversion import convert_names
 
 _IGNORE_FOLDERS = ('.tox', 'venv', '__pycache__')
 
-_OPENED_FILE_LIMIT = 1000
+_OPENED_FILE_LIMIT = 2000
+"""
+Stats from a 2016 Lenovo Notebook running Linux:
+With os.walk, it takes about 10s to scan 11'000 files (without filesystem
+caching). Once cached it only takes 5s. So it is expected that reading all
+those files might take a few seconds, but not a lot more.
+"""
 _PARSED_FILE_LIMIT = 50
+"""
+For now we keep the amount of parsed files really low, since parsing might take
+easily 100ms for bigger files.
+"""
 
 
 def _resolve_names(definition_names, avoid_names=()):