Parallax is one of the easiest effects to prototype and one of the easiest to ruin in production. It takes an afternoon to get a layer moving at a different speed than the page around it. It takes a lot longer to make that feel free on a mid-range Android phone instead of just your MacBook — which is exactly why performance rules like these are baked into Vertex, the platform behind every site we ship.
Where Parallax Actually Breaks
Most parallax implementations bind a scroll listener directly to a transform or top update, then call it a day. On a high-end laptop this looks smooth. On the device your actual customer is holding, the scroll event fires on the main thread, the layout recalculates, and the frame drops the moment anything else — a font loading, an image decoding, a re-render — competes for that same thread.
The gap between “works on my machine” and “works” is almost always a main-thread problem, not a math problem.
The Fix Is Architectural, Not Clever
The first rule: never let scroll position update layout-triggering properties. top, left, width, and height all force the browser to recalculate the page. transform and opacity do not — they run on the compositor thread, separate from the thread doing everything else. This single substitution is responsible for most of the performance gain in any parallax rebuild we do.
The second rule: don't listen to scroll directly. We use GSAP's ScrollTrigger, which batches scroll reads and writes using requestAnimationFrame internally, so the browser is never asked to do more than one layout pass per frame regardless of how fast someone scrolls.
The third rule: pin sparingly. Pinned sections are the most expensive parallax pattern because they usually require recalculating scroll-linked positions for multiple sibling elements at once. We use them where they earn their cost, not by default.
“A parallax section that stutters reads as unfinished, not ambitious.”
Testing on the Right Hardware
A parallax section that hits 60fps in Chrome DevTools on a desktop can drop to 20fps on a three-year-old Android phone with a locked CPU. We test every scroll-driven section on throttled CPU and on at least one real mid-range device before it ships — a standard enforced at the Vertex platform level, not left to individual project discretion.

Compositor-thread animation vs. layout-thread animation
What This Means for the Work
Motion is only a craft signal if it's smooth. A parallax section that stutters reads as unfinished, not ambitious. Every scroll interaction on this site follows the three rules above, because Vertex enforces them by default. It's not a shortcut. It's the actual discipline behind “it just feels right.”

