Optimizing Resource Allocation for cPanel Managed Hosting Services?

We've been running a specialized service providing website maintenance and cPanel management for a diverse set of clients, and while our business is growing, we're consistently hitting a wall with effective resource allocation in our multi-tenant hosting environments. It’s a classic problem, but one that seems to defy truly elegant solutions. Despite implementing CloudLinux with LVE Manager across our cPanel servers, we're still grappling with intermittent yet significant CPU, RAM, and IOPS spikes originating from individual client sites. These resource hogs, even when contained by LVEs to some extent, invariably introduce performance degradation and instability across the entire server, negatively impacting other clients who are well within their allocated limits.

Our current setup, leveraging standard cPanel/WHM configurations augmented by CloudLinux, provides a baseline level of resource isolation, but it's clearly not granular or robust enough for the dynamic and often unpredictable loads our clients generate. We're looking beyond the default LVE settings and basic throttling capabilities because they simply aren't delivering the consistent performance stability and true resource isolation we need for a premium managed service offering. So, my core technical question revolves around what advanced, kernel-level, or containerization-based strategies are proving genuinely effective for isolating and proactively throttling these resource-intensive accounts in a cPanel managed hosting environment. Are there specific, successful implementations of highly tuned cgroup configurations, custom kernel patches, or perhaps even lightweight containerization approaches (beyond what CloudLinux offers out-of-the-box) that provide superior performance stability and more effective resource isolation without completely overhauling the cPanel ecosystem? We’re particularly interested in practical, actionable advice from those who have successfully implemented such deep-level optimizations. Help a brother out please...