I also have performance issues with version 1.7.0.
I created the following patch.
If the node is not cached, getting file size and removing cache of the node would not be required.

--- a/src/cache.c
+++ b/src/cache.c
@@ -1022,7 +1022,8 @@ dav_lookup(dav_node **nodep, dav_node *parent, const char *name, uid_t uid)

         if (is_open(*nodep))
             attr_from_cache_file(*nodep);
-        else if (!is_dirty(*nodep)) {
+        else if (!is_dirty(*nodep) && !is_backup(*nodep) & is_cached(*nodep)
+                     && time(NULL) > ((*nodep)->utime + file_refresh)) {
             int ret;

             dav_props *props = NULL;

The regression is caused by the fix for Bug #61049. I don't know if both the desired caching behaviour and efficient directory traversal can be achieved with the FUSE API, so maybe the user might need to choose between the old and current implementation through a config setting (similar to gui_optimize).

From my laptop, I access a nextcloud instance, which is mounted at the directory ~/nextcloud. Under davfs2 version 1.6.1, the command "ls -l ~/nextcloud/presse" takes about one second. However, using version 1.7.0, the same command takes about 150 seconds (two minutes and a half). The same problem existed already with an earlier version of davfs2 (approximately version 1.5). This problem is fully reproducible.