The traditional fetch() helper methods (response.json(), response.text(), response.blob()) are monolithic buffering operations: they force the browser to buffer the entire HTTP payload into memory before executing the callback. For large payloads (e.g. 500MB CSV files or large GeoJSON datasets), this results in severe RAM spikes, multi-second Time-To-First-Token (TTFT) delays, and frequent browser tab crashes due to out-of-memory (OOM) killer terminations. For real-time LLM inference streams (OpenAI, Anthropic, Ollama), buffering prevents incremental token rendering. The WHATWG Streams API (response.body as ReadableStream) allows applications to process data chunk-by-chunk as individual bytes arrive over the network, achieving sub-millisecond initial display times.

Backpressure is the signaling mechanism by which a slow consumer prevents a fast producer from overflowing system memory. In a streaming pipeline constructed with readable.pipeThrough(transform).pipeTo(writable), each stream maintains an internal queue governed by a highWaterMark strategy. If the consumer (e.g. rendering SVG elements or writing to IndexedDB) falls behind, its queue fills up and its controller.desiredSize drops to 0 or negative. This backpressure automatically propagates backwards through the TransformStream stages to the underlying ReadableStream, which suspends pull() calls on the network socket. This throttles TCP/QUIC flow control windows at the transport layer, capping memory consumption at a constant few kilobytes regardless of total payload size.

Standard default stream readers (readable.getReader()) allocate a fresh Uint8Array chunk inside the browser engine for every incoming network packet. In high-throughput streaming (video transcoding, audio DSP, or high-frequency telemetry), allocating tens of thousands of short-lived typed arrays generates severe garbage collection (GC) thrashing that causes micro-stutters and dropped UI frames. If a ReadableStream uses byte stream mode (type: "bytes"), consumers can call readable.getReader({ mode: "byob" }). BYOB allows the consumer to supply an existing pre-allocated ArrayBuffer slice directly to reader.read(bufferView). The browser writes incoming network bytes directly into that memory, achieving true zero-allocation zero-copy streaming.

The WHATWG Streams specification allows modular stream transformations using pipeThrough(). Because TransformStream instances expose a readable property and a writable property, pipelines can be chained linearly: response.body.pipeThrough(new DecompressionStream("gzip")).pipeThrough(new TextDecoderStream()).pipeThrough(new LineSplitterStream()).pipeThrough(new JsonParserStream()).pipeTo(consumerSink). Each stage encapsulates a single transformation concern while preserving backpressure and error propagation across the entire chain.