From what I understand your site uses a custom-built Drupal integration plugin, so I can't comment on that code having never actually seen it.

Search speed in general is sensitive to many different factors. To start, try searching on the back-end. Is it slow there, or only on the Drupal front-end? If it's also slow on the back-end then it's likely either an issue with the server configuration, database configuration or the selected CA search engine.

If it's only slow on the front-end then it's either probably either a front-end server configuration issue, a network performance issue (assuming the front-end is hitting a database on the back-end machine) or an issue with your Drupal code.

We have encountered search performance issues with large datasets (> 100,000 "heavy" records - those with tons of text and relationships) and have been working to address them. Much of the problem stems from the "refine search" feature - the browsing within search results. There are optimizations in 1.0 that largely address this and there are further refinements in 1.1 that attempt to reduce overhead to nearly zero. If you are using a pre-1.0 version and haven't updated in the last few months you may not have this code yet.

The choice of search engine also makes a big difference. Versions of CA up to 1.0 use an engine based upon the fulltext indexing features in MySQL. We thought this would be a clever choice and perform well, but in in actual use MySQL fulltext support has been shown to have significant performance issues with queries on words that appear often, short words (< 4 characters) and codes that include non-alphanumeric characters (eg. accession numbers).

For 1.1 we have implemented a new search engine that employs an inverted index in a MySQL table. It is zero-configuration like MySQL fulltext, but appears to have improved performance and scaleability.

MySQLFulltext and the inverted index engine in 1.1 are both meant as entry-level engines. You may also want to try the Apache SOLR engine plugin. SOLR is proven to scale very very well and provides great performance on large datasets. But it does require installation and maintenance of a separate pieces of software, and isn't easy to host on a shared server.

I should note that there a many many installations of the same scale as yours (15,000 or so records) that have no performance issues, even when using MySQL Fulltext. I would definitely start by making sure your code is up-to-date, that the servers you are running on have enough memory (4 gigs or more is generally good), that the MySQL configuration in your my.cnf file is sensible and that there is good network connectivity between the back-end database and front-end web sites.

seth